2022 44th Annual International Conference of
the IEEE Engineering in Medicine & Biology Society (EMBC)
Scottish Event Campus, Glasgow, UK, July 11-15, 2022
2022 44th Annual International Conference of the IEEE Engineering in Medicine & Biology Society (EMBC) | 978-1-7281-2782-8/22/$31.00 ©2022 IEEE | DOI: 10.1109/EMBC48229.2022.9870910
Abdominal cardiovascular sound recording and analysis using
cardio-microphones
Julie Fontecave-Jallon, Amira Haouas, Stéphane Tanguy
Abstract— In view of using abdominal microphones for fetal
heart rate (FHR) monitoring, the analysis of the obtained
abdominal phonocardiogram (PCG) signals is complex due to
many interferential noises including blood flow sounds.
In order to improve the understanding of abdominal
phonocardiography, a preliminary study was conducted in one
healthy volunteer and designed to characterize the PCG signals all
over the abdomen. Acquisitions of PCG signals in different
abdominal areas were realized, synchronously with one thoracic
PCG signal and one electrocardiogram signal. The analysis was
carried out based on the temporal behavior, amplitude and mean
pattern of each signal.
The synchronized rhythmic signature of each signal confirms that
the PCG signals obtained on the abdominal area are resulting
from heart function. However, the abdominal PCG patterns are
totally different from the thoracic PCG one, suggesting the
recording of vascular blood flow sounds on the abdomen instead
of cardiac valve sounds. Moreover, the abdominal signal
magnitude depends on the sensor position and therefore to the size
of the underlying vessel.
The sounds characterization of abdominal PCG signals could
help improving the processing of such signals in the purpose of
FHR monitoring.
I. INTRODUCTION
For decades, phonoangiography, i.e. recording of sounds
generated by blood flow in vessels, is described as a simple
examination that can be used to determine flow perturbations
in pathophysiological situations including atherosclerosis [1].
This technique using microphones is still under investigation,
notably in order to create new medical devices designed to
increase the accuracy of the sound analysis [2]. When focusing
on a specific area (e.g. carotid area), the signals are well
described and can be both qualitatively and quantitatively
analyzed. However, on larger area such as the abdominal area,
this auscultation is made difficult by the numerous vessels that
are present. The resulting recordings of blow flood sounds from
all those vessels, lead to signals, whose pattern and magnitude
will differ according to the sensor positions.
Analyzing the fetal heart rate (FHR) and its variability is
used to follow the fetal well-being during pregnancy. Among
the FHR monitoring non-invasive techniques currently in
development, the use of cardio-microphones is of interest [3-
5]. Usually, cardio-microphones are placed on the thoracic
This work was partially supported by the French National Research
Agency, as part of the SurFAO project (ANR-17-CE19-0012). Julie
Fontecave-Jallon, Amira Haouas and Stephane Tanguy are with Univ.
978-1-7281-2782-8/22/$31.00 ©2022 IEEE
Authorized licensed use limited to: Terna Engineering College - Navi Mumbai. Downloaded on November 09,2022 at 08:13:27 UTC from IEEE Xplore. Restrictions apply.
area [6] and give phonocardiogram (PCG) signals, related to
heart sounds (cardiac valves opening and closure). In the
context of FHR monitoring, these sensors are positioned on the
mother’s abdomen to record fetal sounds [5]. The analysis of
the obtained abdominal PCG signals is complex due to many
interferential noises, potentially including the above-mentioned
maternal blood flow sounds.
In this context, we propose a preliminary study in a non-
pregnant healthy volunteer, designed to characterize the PCG
signals obtained in the different abdominal areas. We aim to
further use this characterization for FHR monitoring
investigation.
II. MATERIAL AND METHODS
A. Data acquisition
Acquisitions of abdominal and thoracic PCG signals,
simultaneously with an electrocardiogram (ECG) signal were
performed on one healthy female volunteer (22 years old) in a
dedicated place of biomedical research at TIMC laboratory (La
Tronche, France), using a PowerLab data acquisition system
(ADInstruments). All signals were sampled at 1 KHz.
Measurements were performed in a quiet laboratory (without
special acoustic shielding) in sitting position. Once the
experimental design was properly defined, one final 1-minute
acquisition was realized and considered for analysis.
The subject was equipped with 3 disposable electrodes
(Ag/AgCl) for a DII lead ECG and 15 cardiac microphones
(MLT201, AD Instruments) put on the skin for PCG
acquisitions. One microphone was placed in front of the heart
and provided a thoracic PCG signal, noted as PCGT. For
abdominal PCG, 15 abdominal auscultation sites under the
umbilic were considered, as shown on Figure 1. The abdomen
was regularly sampled with 3 lines (i = 1 to 3, from top to
bottom) and 5 columns (j = 1 to 5, from left to right). Each
corresponding abdominal PCG signal is noted as PCGAij.
Microphones were fixed on the chest and abdomen using
Tegaderm (3M) transparent medical dressing. The A13 position
corresponding to the umbilicus has not been considered due to
microphone fixation.
Grenoble Alpes, CNRS, CHU Grenoble Alpes, Grenoble INP, TIMC,
Grenoble, France (phone: +33456520063; fax: +33456520033; e-mail:
julie.fontecave@univ-grenoble-alpes.fr).
820
 Figure 1. Abdominal auscultation sites for abdominal PCG recordings. 15
poisitons are considered to sample the abdomen under the umbilic, these
positions are noted Aij where i denotes the line and j the column. The A13
(black disk) recording has not been performed.
B. Signal processing
All processing steps were carried out using Python/Spyder.
1) Signal filtering
ECG signals were first processed through a Finite Impulse
Response (FIR) band-pass filter [2-40] Hz (order 200) so as to
enhance the R-peaks.
As usually considered for cardiac sound analysis, thoracic
PCG signals PCGT were band-pass filtered between 15 and 150
Hz (FIR filter, order 200) [6]. This allows to highlight (as
shown on Fig. 2, middle panel) the more audible heart sounds,
S1 and S2, which correspond to the closure of respectively the
atrial-ventricular valves (beginning of the ventricular systole)
and the aortic and pulmonary valves (onset of the ventricular
diastole). These 2 sounds are systematically following R-peaks
of synchronous ECG signal.
However, when analyzing the Power Spectral Densities
(PSD) of raw abdominal PCG signals, one can notice that the
spectral bandwidth is much lower than for raw thoracic PCG.
Whatever the sensor position, most of abdominal PCG energy
is lower than 25 Hz, as observed for one PCGA on Fig.3. The
band-pass filter considered for PCGT is then not adequate for
abdominal PCG. Therefore, each abdominal PCG signal PCGA
was band-pass filtered between 5 and 20 Hz (FIR filter, order
300). One example can be seen for one position on Fig.2
(bottom panel). This underlines the repetition of one pattern at
each heart beat characterized by the ECG R-peaks.
Figure 2. Synchronous band-pass filtered ECG, thoracic PCG and A23
abdominal PCG. Band-pass filtering is different for each signal ([2-40] Hz for
ECG, [15-150] Hz for PCGT and [5-20] Hz for PCGA). Amplitudes are
expressed in mV.
821
Authorized licensed use limited to: Terna Engineering College - Navi Mumbai. Downloaded on November 09,2022 at 08:13:27 UTC from IEEE Xplore. Restrictions apply.
Figure 3. Power Spectral Density of thoracic and abdominal PCG signals.
2) Mean patterns of PCG signals from ECG R-peaks
detection
In order to analyze the cardiovascular sounds recorded by
PCG sensors both on thoracic and abdominal positions, mean
patterns of PCG signals were computed according to ECG R-
peaks locations.
First, R-peaks were detected on each filtered ECG signal
with Pan & Tomkins’ well-known algorithm [7]. Then, ECG
signals were segmented according to these R-peaks locations.
For each heartbeat, the segmentation window started 200ms
before the R-peak and finished 650ms after the R-peak, as
illustrated on Fig.4a. This allows to consider the whole
cardiac cycle, with P, Q, R, S, T waves. Windows were not
centered on R-peaks for a better visualization of
cardiovascular sounds on PCG signals, as it will be observed
below. ECG segments were then superimposed (Fig. 4b) and
averaged (Fig.4c) so as to obtain the mean ECG pattern. 50
segments were considered in the average computation.
The same process was applied to PCG signals. According
to R-peaks previous locations, PCG signals were segmented
with the same window as the one defined for ECG. PCG
segments were superimposed and then averaged.
Figure 5 illustrates the process for both thoracic and
abdominal PCG (one example in one sensor position). Fig. 5a
and Fig. 5c respectively show the superimposition of band-
pass filtered thoracic and abdominal PCG. Fig. 5b and Fig. 5d
show the mean patterns of thoracic and abdominal PCG.
Figure 4. Ensemble avergae of ECG signals. (a) R-peak detection and ECG
segmentation, (b) superimposition of ECG segments (c) mean pattern of ECG
signal.
 Figure 5. Ensemble average of thoracic and abdominal PCG signals. Pattern
durations = 850 ms. (a) superimposition of thoracic PCG segments (b) mean
pattern of thoraic PCG signal, (c) superimposition of abdominal PCG
segments, (d) mean pattern of abdominal PCG signal.
III. RESULTS
A. Temporal behavior of abdominal PCG signals
As previously mentioned and as it can be observed on Fig.
2, the temporal analysis of signals highlights a quasi-periodic
behavior of each signal. For each cardiac cycle underlined by
the R-peaks of the ECG, one can see a pair of 2 cardiac sounds
S1 and S2 on thoracic PCG signal and a specific pattern on
abdominal PCG.
Considering each position of abdominal PCG sensors, Fig. 6
shows the abdominal PCG rhythmic patterns for a few cardiac
cycles. Whatever the position, these patterns are close one to
each other but very different from the S1 and S2 sounds of
opening and closure of cardiac valves (Fig. 2, middle panel).
The same representation is realized on Fig. 7 but with
amplitudes normalized on each subplot. It appears that there is
a high variability of PCG signal amplitudes depending on the
sensor position.
Figure 6. Temporal respresentation of band-pass filtered abdominal PCGAij
signals during 3.5 sec (i = 1 to 3, from top to bottom, and j = 1 to 5, from left
to right). Amplitude in mV is not normalized.
822
Authorized licensed use limited to: Terna Engineering College - Navi Mumbai. Downloaded on November 09,2022 at 08:13:27 UTC from IEEE Xplore. Restrictions apply.
Figure 7. Temporal respresentation of band-pass filtered abdominal PCGAij
signals during 3.5 sec (i = 1 to 3, from top to bottom, and j = 1 to 5, from left
to right). Amplitude has been normalized between [-20,20] mV.
B. Mean pattern of abdominal PCG signals
In the same way, Figure 8 represents the abdominal PCG
mean patterns according to the sensor positions. Amplitude has
been normalized for display. Again, the amplitude disparities
can be observed with small signals for some positions and
much higher amplitudes for others. Moreover, we observe that
the mean patterns look very similar in each position, and
completely different from the thoracic PCG mean pattern
(Fig.5b).
Figure 8. Mean pattern of band-pass filtered abdominal PCGAij signals
during 3.5 sec (i = 1 to 3, from top to bottom, and j = 1 to 5, from left to right).
Amplitude hab been normalized between [-20,20] mV.
IV. CONCLUSION
Although the dataset obtained on a single volunteer is
limited, the present results make it possible to advance in the
analysis of abdominal phonocardiography.
Based on the synchronized rhythmic signature of the
thoracic and abdominal PCG obtained in our study, we can
first confirm that the PCG signals obtained on the abdominal
area are resulting from heart function.
The shape difference between thoracic and abdominal
signals is clearly coming from the type of the recorded
 sounds. Indeed, thoracic PCG signals exhibit the well-
described characteristics of sounds generated by the valves of
the heart. When focusing on abdominal PCG, two conclusions
can be drawn. First, the abdominal PCG patterns are totally
different from the thoracic PCG one, suggesting the recording
of vascular blood flow sounds. Secondly, whatever the
positions on the abdomen, the mean PCG patterns remain
similar. However, there is a difference in the signal magnitude
depending on the sensor position and therefore to the
underlying vessel. Indeed, the central (Fig.1, A23) and the low
lateral (Fig.1, A31 and A35) positions of the sensor are
responsible for high amplitude PCG signals. This is in
accordance with the large underlying vessels of these areas,
respectively the aorta and the two femoral arteries, and can be
related to the high blood flow in those important vessels.
In the context of the FHR monitoring with abdominal
cardio-microphones, PCG signals of pregnant women will be
a mixture of both maternal vascular blood flow and cardiac
fetal sounds. According to this preliminary study, fetal PCG
investigation may include the suppression of maternal
vascular sounds for a robust FHR monitoring.

REFERENCES
[1] R.S. Lees, “Phonoangiography: Qualitative and quantitative,” Ann
Biomed Eng 12, 55–62,1984.
[2] T. Sühn, M. Spiller, R. Salvi, S. Hellwig, A. Boese, A. Illanes, M. Friebe,
“Auscultation System for Acquisition of Vascular Sounds -Towards
Sound-Based Monitoring of the Carotid Artery,” Medical Devices:
Evidence and Research, 349-364, 2020.
[3] F. Kovcs, C. Horvth, T. Balogh, G. Hossz, “Fetal phonocardiography past
and future possibilities”, Comput Methods Programs Biomed, Vol. 104,
2011.
[4] S. Gobillot, J. Fontecave-Jallon, V. Equy, B. Rivet, P.Y. Gumery, P.
Hoffmann, “Non-invasive fetal monitoring using electrocardiography and
phonocardiography: a preliminary study”, Journal of Gynecology
Obstetrics and Human Reproduction., 2018.
[5] N. Dia, J. Fontecave-Jallon, P.-Y. Gumery, B. Rivet, "Fetal heart rate
estimation from a single phonocardiogram signal using non-negative
matrix factorization," 2019 41st Ann Int Conf of the IEEE Engineering in
Medicine and Biology Society (EMBC), pp. 5983-5986, 2019.
[6] J. Fontecave-Jallon, K. Fojtik, B. Rivet, "Is there an Optimal Localization
of Cardio-microphone Sensors for Phonocardiogram Analysis?”, 2019
41st Ann Int Conf of the IEEE Engineering in Medicine and Biology
Society (EMBC), pp. 3249-3252, 2019.
[7] J. Pan, W. J. Tompkins, “A Real-Time QRS Detection Algorithm”, IEEE
Trans. Biomed. Eng., vol. BME-32, no 3, p. 230‑236, 1985.

823

Authorized licensed use limited to: Terna Engineering College - Navi Mumbai. Downloaded on November 09,2022 at 08:13:27 UTC from IEEE Xplore. Restrictions apply.